Recent advances in artificial intelligence (AI) have accelerated the development of agentic and multimodal systems, with applications across sectors including healthcare, research, industry and environmental monitoring. As AI capabilities become increasingly integrated into digital products and services, their impact on society continues to grow. While these technologies offer significant opportunities, they also raise important questions regarding transparency, reliability and accountability. Ensuring that AI systems are trustworthy and aligned with societal values remains a key priority for policymakers, researchers and industry.
Recognising this challenge, the European approach to Artificial Intelligence emphasises that trust is essential and risks related to this technology need to be considered early on. As a guiding principle, accountability, transparency, and fairness need to be included by design. In this context, HaDEA will be present at the International Conference on Engineering, Technology and Innovation (ICE2026), presenting its activities in a keynote talk. Karina Marcus, from the Horizon Europe: Digital team at HaDEA, will also chair a workshop titled “Robust and trustworthy AI is not one-size-fits-all”, where six HaDEA-funded projects will share their experience and discuss how to translate high-level principles of trustworthy AI into concrete methods, tools, and decisions in practice. Meet the six projects that will participate and contribute to the workshop.
- AIXPERT develops a novel approach to developing AI systems that are explainable, accountable, and transparent. The project is centred on an adaptable, situation-aware AI-agentic platform capable of encapsulating various AI models, regardless of their underlying architecture. This approach significantly enhances the trustworthiness of AI systems by providing a consistent framework for explainability and accountability across different model types. The project tackles challenges in making AI systems explainable, transparent, accountable, autonomous, and robust by integrating multi-agent systems with foundation models that use multiple forms of data and real-time human feedback. This combination enhances AI system trustworthiness and user-friendliness while allowing for flexibility in the choice of underlying AI models. The project’s results will be applied in five specific areas: healthcare, recruitment, manufacturing, education, and the creative industries.
- EXTRA-BRAIN develops next-generation explainable and trustworthy AI, inspired by how the human brain processes information, moving beyond traditional deep neural networks that rely on resource-intensive learning processes using global learning rules. Instead, it adopts brain-like mechanisms such as selective attention, parallel and asynchronous processing, adaptability, and efficient resource use. By leveraging insights from computational neuroscience, the project aims to create scalable, energy-efficient, and flexible brain-like neural networks that can learn from a few examples. Its main goal is to bring these advanced approaches out of research and deploy them in real-world applications. The project outcomes will be validated in three different scenarios, including perception in autonomous robots, efficient (both in terms of data and power) AI for digital finance and optimal resource allocation for end users in telecommunication networks.
- FAITH develops a novel human-centric, holistic AI trustworthiness assessment framework that has the potential to support the entire AI system development lifecycle, including data preparation, algorithmic design, development, and deployment, as well as operation, monitoring, and governance. Moreover, one of the key ambitions of the project is to orient various stakeholders towards an AI paradigm shift in terms of trustworthiness, from current rather fragmented to a systematic trustworthiness management ecosystem approach. The project also establishes an AI trustworthiness evaluation approach that will support the transfer of knowledge from one sector into another through a two-phase evaluation process. The project outcomes will be tested on seven critical domains including robotics, education, media, transportation, healthcare, active ageing, and industry.
- HumAIne develops a novel AI paradigm to foster trusted human-AI collaboration in dynamic, unstructured environments, leveraging active learning techniques and neuro-symbolic AI. The project proposes a novel standards-based reference architecture that will support human-centric, safe, trustworthy and regulatory-compliant human-AI use case implementation by design. Moreover, the novel approach will provide semantic interoperability and knowledge sharing between humans and AI, which is a precondition for interpretability, transparency and explainability of different AI models that are provided to the users. The system will also address the problem of uncertainty, which is inherent to interpreting data in different contexts, as well as the training, execution, and combination of intelligence in federated AI systems. The project outcomes will be tested in three domains, including healthcare, precision farming and industry.
- TRUMAN develops generic technologies and methods to improve the resilience of AI systems against security, privacy, and fairness threats, and to increase user trust in these systems. This will be done by considering all stages of the AI life cycle, from data collection to training and deployment. The project also develops customised solutions to protect against existing and new types of attacks, including those that target privacy, fairness, and system security. Moreover, the project examines the impact of these solutions on people and develops ways to explain the underlying technologies and risks to them, as well as involve them in improving these models. The project will also work to establish comprehensive guidelines and recommendations for building trustworthy and robust AI systems. The project’s technologies will be tested on four real-world scenarios from different sectors: marketing, IT security, finance, and healthcare, to evaluate their effectiveness.
- TURING develops novel physics-aware generative AI framework that includes multimodal foundation models, as well as task-specific models, that can accurately capture the physical properties of complex systems. Such AI systems are widely used to model and simulate real physical phenomena, e.g. turbulence, weather patterns, ocean currents or planetary formation. The novel system will exhibit advanced generalisation capabilities, while maintaining robustness and reliability for a broad range of user-specified problems. The novel approach will also address the problem of uncertainty that is a precondition for interpretable results compared to existing black-box AI models. Moreover, the novel approach will develop meta-learning models that can adapt to new tasks from different application domains quickly, or with low data and computational effort. The project results will be tested in three areas, including nuclear energy, high-energy physics, and meteorology.